Kubernetes Architecture and Components¶
Overview¶
Trace a request from kubectl through the API server to etcd, scheduling, and the kubelet — and name every major control-plane piece.
The control plane stores and reconciles desired state. Workers run Pods. Everything goes through the Kubernetes API.
This is a core tutorial in Module 1 · Kubernetes Fundamentals of the REBASH Academy Kubernetes for Cloud & DevOps Engineers series — written for Cloud, DevOps, Platform, and SRE engineers.
Prerequisites¶
Learning Objectives¶
By the end of this tutorial, you will be able to:
- List API server, etcd, scheduler, controller-manager
- Explain kubelet and kube-proxy roles
- Describe cloud-controller-manager (when present)
- Relate CNI to Pod networking
Architecture¶
This topic’s control points and relationships are shown below.
Theory¶
What it is¶
A Kubernetes cluster has two logical halves: the control plane (brains) and worker nodes (muscle). Everything you do with kubectl is an API call against the control plane. Workers run your Pods under instructions from that API. Understanding the components lets you debug “who failed?” instead of guessing at YAML.
Why it matters¶
When a Deployment will not schedule, a node goes NotReady, or kubectl hangs, you need to know whether the API server, etcd, the scheduler, or the kubelet is involved. Architecture knowledge is the difference between productive troubleshooting and random restarts. Certifications (CKA/KCNA) and on-call work both assume this map.
How it works (mental model)¶
Trace a typical create:
kubectl apply→ kube-apiserver (authentication, authorisation, admission).- Object persisted in etcd.
- Controllers in kube-controller-manager notice desired replicas and create Pods (still unbound).
- kube-scheduler writes a
nodeNamebinding on the Pod. - The kubelet on that node asks the container runtime (via CRI — Container Runtime Interface) to pull and start containers.
- kube-proxy (or an eBPF dataplane) programmes rules so Services reach Pod IPs.
Reconciliation never stops: controllers and kubelets continuously compare desired vs actual state.
Key concepts / comparisons¶
| Component | Role |
|---|---|
| kube-apiserver | Front door; authn/authz |
| etcd | Cluster state store |
| kube-scheduler | Assign Pods to nodes |
| kube-controller-manager | Deployments, Nodes, endpoints, etc. |
| cloud-controller-manager | Cloud LB/volumes/routes (when present) |
| kubelet | Node agent; runs Pods |
| kube-proxy | Service dataplane (often) |
| container runtime | containerd / CRI-O |
| CNI plugin | Pod network addresses and routes |
Control plane may run on dedicated machines or as managed services (EKS/AKS/GKE). Workers hold application Pods. High availability means multiple API server and etcd members — a single control-plane VM is a lab pattern, not a production pattern.
Common pitfalls¶
- Assuming etcd is “just a database” — corrupting or starving etcd takes the cluster down.
- Forgetting that kube-proxy is not the only Service implementation; Cilium and others may replace it.
- Expecting
kubectl get componentstatusesto be the modern health check — prefer node Ready, API reachability, and control-plane Pods inkube-system. - Confusing CNI (Pod networking) with kube-proxy (Service VIP distribution).
Hands-on Lab¶
Create a workspace for this tutorial.
Focus: Map control plane and node components from a live cluster
Step 1 – Inspect nodes and system Pods¶
kubectl create namespace rebash-lab
kubectl get nodes -o custom-columns=NAME:.metadata.name,ROLES:.metadata.labels.node-role\.kubernetes\.io/control-plane,VERSION:.status.nodeInfo.kubeletVersion
kubectl get pods -n kube-system -o wide
kubectl get --raw /readyz?verbose | head -n 20 || true
Step 2 – Identify API server and scheduling signals¶
kubectl cluster-info
kubectl get componentstatuses 2>/dev/null || kubectl get --raw /livez | head -c 200; echo
kubectl explain pod.spec.nodeName
kubectl -n rebash-lab run arch-probe --image=busybox:1.36 --restart=Never --command -- sleep 30
kubectl -n rebash-lab get pod arch-probe -o jsonpath='Node={.spec.nodeName} Phase={.status.phase}{"
"}'
Final step – Cleanup note¶
kubectl delete namespace rebash-lab --ignore-not-found
# Workspace kept for notes; remove with: rm -rf "$(pwd)" when finished
Validation¶
- Lab commands run under
~/rebash-k8s/module-01-arch/ - You can explain each Theory section in your own words
- You used modern tooling where it applies to this topic
- You can describe one production failure mode for this topic
Code Walkthrough¶
Production practice for Kubernetes Architecture and Components always combines:
- Inspect before you change (status, plan, logs, dry-run)
- Prefer reversible, documented changes (Git, IaC, drop-ins, version pins)
- Capture evidence (command output, pipeline logs) for handovers
- Prefer current tools and APIs over legacy shortcuts
- Least privilege — escalate credentials only when required
Keep runbooks short enough to follow under pressure. Automate checks; keep humans for judgement.
Security Considerations¶
- Treat credentials and tokens for kubernetes as privileged — never commit them
- Prefer short-lived auth (OIDC, roles, SSO) over long-lived keys
- Validate blast radius before apply/deploy/delete operations
- Restrict who can approve production changes
- Collect audit logs; limit who can read sensitive traces
Common Mistakes¶
Assuming etcd is “just a database” — corrupting or starving etcd takes the cluster down.
Validate assumptions against the Theory section and official docs before changing production.
Forgetting that kube-proxy is not the only Service implementation; Cilium and others may r
Lab shortcuts (open security groups, admin roles, skip approvals) must not ship unchanged.
Changing production without a rollback path
Always know how to revert (previous artefact, prior release, state rollback, DNS failback).
Best Practices¶
- Encode Kubernetes Architecture and Components changes as code and review them in pull requests
- Pin versions (images, modules, actions, provider plugins)
- Separate environments with clear promotion gates
- Alert on symptoms with runbooks attached
- Destroy lab resources; tag everything with owner and expiry where possible
Troubleshooting¶
| Symptom | Likely cause | Fix |
|---|---|---|
| Auth / permission denied | Wrong identity, policy, or scope | Check caller identity, roles, and least-privilege policies |
| Timeout / no route | Network, DNS, security group, or endpoint | Trace path, DNS, and allow-lists before retrying |
| Drift / unexpected plan | Manual change or wrong state/workspace | Reconcile desired vs actual; avoid click-ops on managed resources |
| Pipeline/job red | Flaky step, cache, or missing secret | Read failing step logs; bisect recent workflow/config changes |
| Cost spike | Idle load balancer, NAT, oversized compute | Inventory billable resources; stop/delete labs promptly |
Summary¶
Kubernetes Architecture and Components is essential for Cloud and DevOps engineers working with kubernetes. Practise the lab until the inspection and change path is muscle memory, then continue the track.
Interview Questions¶
- What are the main control plane components of Kubernetes?
- What is the role of kubelet versus kube-proxy on a node?
- Where does etcd fit, and why is its health critical?
- How does a highly available control plane change failure domains compared with a single API server?
- What is the scheduler responsible for?
Sample answer — question 2
kubelet ensures Pod specs assigned to the node are running and reports status. kube-proxy programmes Service networking rules (or relies on equivalent dataplane) so ClusterIP traffic reaches Pods.
Sample answer — question 4
Multiple API server and etcd members reduce single points of failure, but you must still plan for quorum, load balancing, and zone-aware placement so correlated failures do not take the whole control plane down.